Predicting the dynamics of an oligo-oscillatory reaction by an artificial neural network

Artificial neural networks (ANNs) are model-free computationat tools dtat can "learn" dte linear or nonlinear rules embedded in a dataset. We report dte results of an attempt to utilize ANNs in dte field of reaction kinetics. A feedforward network is trained to predict dte main dynamical feamres of oligoasci/lations in dte acidic bromate-ascorbic acid-malonic acid reacting mixture, in which the concentration of bromide ion (an intermediate) shows three extrema as a function of time. Inputs to die network are the initial concentrations of reactants, while outputs are dte predicted values of bromide-ion co~ntration and reaction time at dte extrema. The network is first tested on a numerically generated dataset uxI dten applied to experiments. The results provide evidence dtat ANNs can be efficiently employed for the prediction of the dynamics of complex chemical systems, especially, when the mechanism of a reaction is not fully understood.